Accelerating time-to-treatment in rural areas with a pre-hospital EEG stroke classification system - stroke is the second leading cause of death and the primary cause of long-term disability worldwide…

The National Institute for Minority Health and Health Disparities awarded Asterion AI Inc. $314,363 on August 5, 2026, under the Minority Health and Health Disparities Research program (CFDA 93.307) to develop and validate a pre-hospital electroencephalogram-based stroke classification system for rural emergency response.

Asterion AI will deploy a CE-marked, FDA-cleared EEG system paired with machine learning models to differentiate ischemic from hemorrhagic stroke in the pre-hospital setting, operating within or alongside mobile stroke units and standard EMS. The system autonomously eliminates EEG artifacts and provides real-time stroke classification to accelerate treatment initiation. The underlying models achieve stroke-versus-non-stroke classification at an area under the curve (AUC) greater than 0.95 and large vessel occlusion detection at an AUC of 0.90. By enabling rapid ischemic/hemorrhagic differentiation before hospital arrival, the solution addresses the critical disparity in rural stroke outcomes: rural stroke patients are 50 percent less likely to receive intravenous thrombolysis and experience 8 percent higher mortality than urban patients, partly due to only 26 percent of rural residents living within a 30-minute drive of a certified stroke center.

The grant supports Phase I research and development. Performance occurs in Dallas, Texas, with a period of performance through July 31, 2027.

Accelerating time-to-treatment in rural areas with a pre-hospital EEG stroke classification system - stroke is the second leading cause of death and the primary cause of long-term disability worldwide costing the US $65B every year. A study of nearly 7,000 stroke patients revealed that the average onset-to-intervention time was nearly 4HRS, with the best outcomes achieved under three hours and statistically significant improvements for each 15-minute window under that threshold. Timely interventions can reduce long-term disability, but not all interventions work for different types of stroke. 87% of strokes are ischemic in nature (AIS), but administering anti-coagulants to the 13% of hemorrhagic stroke patients would be disastrous. Because current EMS capabilities cannot reliably differentiate between ischemic and hemorrhagic stroke before hospital arrival, very few are treated with TPA within the critical first hour. Disparities further widen in rural areas, where stroke patients are 50% less likely to receive intravenous thrombolysis and have 8% higher mortality than urban patients. Only 26% of rural residents live within a 30-minute drive of a certified stroke center (versus 70% of urban residents) making intervention during the "golden hour" highly unlikely. To address this GAP, this project will use a CE marked and FDA-cleared EEG system to record and analyze EEG signals to provide a real-time stroke classification solution that works within or alongside mobile stroke units (MSUS) and standard EMS. This is paired with a machine learning model that autonomously eliminates artifacts and accurately triages suspected stroke patients for the fast identification of ischemic/hemorrhagic stroke in the pre-hospital setting. Our product was initially created to detect large vessel occlusion for optimal routing of patients to comprehensive stroke centers for endovascular therapy. Stroke VS non-stroke models operate at an AUC >0.95 and LVO VS non-LVO an AUC of 0.90. Additional work is being conducted with the LOS angeles county emergency medical services agency. This phase I project pursues three specific aims. 1) multi-site clinical validation of EEG-based detection of hemorrhagic stroke. Data will be collected from EMS environments to ensure model performance in real-world settings. The target is an AUC ≥0.90 for ischemic VS. Hemorrhagic classification. Secondary goals include sensitivity and specificity ≥85% with special emphasis on achieving high hemorrhagic sensitivity to support safe TPA decisions in prehospital care. 2) ensure EEG signal robustness across diverse hair types. Previous studies have shown that certain phenotypes pose challenges for electrode recordings. The goal is to demonstrate non disparity between hair type subgroups with…